N環球 N GLOBE
US Technology

Nvidia's next-gen optics-on-interposer 'OOI' surfaces: optical engines move onto the silicon interposer, but it is still only a prototype

· N GLOBE

DIGITIMES reported on October 25 that Nvidia is developing next-generation co-packaged optics, codenamed OOI, moving optical engines from the organic package substrate onto the silicon interposer to shorten electrical paths, raise bandwidth density and cut power. It remains a prototype-stage research effort, with no GPU product adoption or volume production announced.

DIGITIMES reported on October 25 that Nvidia is developing next-generation co-packaged optics (CPO) technology, codenamed OOI, short for Optics-on-Interposer. The design moves the optical engines that convert signals between electrical and optical form from the conventional organic package substrate onto the silicon interposer, aiming to shorten the electrical path from the compute chip, raise bandwidth density and lower power.

What changes when optics move onto the interposer

Inside chips such as GPUs, signals are essentially electrical, so optical links need optical engines to convert between the two. Conventional CPO places those engines on an organic package substrate; OOI puts them on the silicon interposer instead. The interposer's fine wiring can be denser and the electrical path shorter, which in theory helps bandwidth density and power.

A shorter path does not automatically mean a large cut in total system power, however. The real benefit depends on the optical engines themselves, the electrical circuitry, thermal design and package structure: the area around the compute chip runs hot, so the temperature behaviour and reliability of optical devices matter, and when a GPU and optical engines share one expensive package, a defect in either one has a significant effect on manufacturing cost.

Still a prototype, not a product

The key distinction is that OOI remains at the research and prototype stage. Nvidia has not announced its adoption in any specific GPU product or the start of volume production, and questions remain over assembly flow, test methods, yield and how failures are handled. Contemporary coverage likewise cautions that a prototype cannot be equated with orders or revenue already booked.

The work is part of Nvidia's push to integrate optics more tightly into AI data-centre architecture. It is a separate development from the DSX POD modular data-centre design this site covered on October 21. Whether OOI moves from prototype to production will depend on future product announcements and customer adoption.

💬 Comments

No comments yet. Be the first!